Structural and Functional Abnormalities of the PrimarySomatosensory Cortex in Diabetic Peripheral Neuropathy:A Multimodal MRI StudyDinesh Selvarajah,1 Iain D. Wilkinson,2 Fang Fang,3 Adithya Sankar,2 Jennifer Davies,2 Elaine Boland,2

Joseph Harding,1 Ganesh Rao,3 Rajiv Gandhi,4 Irene Tracey,5 and Solomon Tesfaye4

Diabetes 2019;68:796–806 | https://doi.org/10.2337/db18-0509

Diabetic distal symmetrical peripheral polyneuropathy(DSP) results in decreased somatosensory cortical graymatter volume, indicating that the disease process mayproduce morphological changes in the brains of thoseaffected. However, no study has examined whetherchanges in brain volume alter the functional organizationof the somatosensory cortex and how this relates to thevarious painful DSP clinical phenotypes. In this case-controlled, multimodal brain MRI study of 44 carefullyphenotyped subjects, we found significant anatomicaland functional changes in the somatosensory cortex.Subjects with painful DSP insensate have the lowestsomatosensory cortical thickness, with expansion ofthe area representing pain in the lower limb to includeface and lip regions. Furthermore, there was a significantrelationship between anatomical and functional changeswithin the somatosensory cortex and severity of theperipheral neuropathy. These data suggest a dynamicplasticity of the brain in DSP driven by the neuropathicprocess. It demonstrates, for the first time in our knowl-edge, a pathophysiological relationship between a clini-cally painful DSP phenotype and alterations in thesomatosensory cortex.

Distal symmetrical peripheral polyneuropathy (DSP)develops in up to 50% of patients with diabetes (1).Most patients develop a painless neuropathy thatincreases the risk of foot ulceration and subsequent am-putation. A significant proportion of patients also develop

a chronic painful condition that can result in considerabledisability and suffering (2,3). Detailed sensory phenotyp-ing has shown distinct sensory profiles of patients withpainful compared with painless DSP (4). Moreover, thesensory profiles of patients with painful DSP were nothomogeneous, with a significant proportion of patientsdemonstrating positive sensory signs, such as dynamicmechanical allodynia (4). No clear pathophysiological ex-planation exists for this spectrum of neuropathic pheno-types in DSP. Much of the research focus has been on theperipheral nervous system only, with potential centralnervous system (CNS) involvement largely overlooked.Although some evidence for CNS involvement has emergedrecently (5–7), further research of the extent of this in-volvement may be crucial for a greater understanding ofthe pathologic mechanisms of DSP, now made possiblewith advances in noninvasive MRI (8).

Structural and functional cortical plasticity is a fun-damental property of the human CNS that enablesadjustment to nerve injury (9,10). However, it canhave maladaptive consequences, possibly resulting inchronic pain (11–13). We have previously demonstrateda clear reduction in both spinal cord cross-sectional area(14,15) and primary somatosensory cortex (S1) (16) graymatter volume in DSP. Studies in other pain conditionsalso have reported dynamic structural and functionalplasticity that have profound effects on the brain inpatients with neuropathic pain (17–19). Collectively, thesestudies led us to hypothesize that both structural and

1Department of Human Metabolism, University of Sheffield, Sheffield, U.K.2Academic Unit of Radiology, University of Sheffield, Sheffield, U.K.3Department of Neurophysiology, Sheffield Teaching Hospitals NHS FoundationTrust, Sheffield, U.K.4Diabetes Research Department, Sheffield Teaching Hospitals NHS FoundationTrust, Sheffield, U.K.5Oxford Centre for Functional MRI of the Brain (FMRIB), Oxford University, Oxford,U.K.

Corresponding author: Dinesh Selvarajah, d.selvarajah@sheffield.ac.uk

Received 13 June 2018 and accepted 13 December 2018

This article contains Supplementary Data online at http://diabetes.diabetesjournals.org/lookup/suppl/doi:10.2337/db18-0509/-/DC1.

© 2019 by the American Diabetes Association. Readers may use this article aslong as the work is properly cited, the use is educational and not for profit, and thework is not altered. More information is available at http://www.diabetesjournals.org/content/license.

796 Diabetes Volume 68, April 2019

COMPLIC

ATIO

NS

functional brain plasticity underly the various clinical phe-notypes of DSP. Hence, the principal aim of this study wasto use combined, structural, and functional MRI (fMRI) toexamine the organization of the S1 in DSP. To our knowl-edge, no study has examined this previously in DSP. If DSPis associated with cortical plasticity, then these changesmay ultimately determine the clinical phenotype.

RESEARCH DESIGN AND METHODS

The study included 44 right-handed subjects (35 with type1 diabetes and 9 healthy volunteers [HVs]). Inclusion cri-teria were type 1 diabetes for .5 years, right-handedness,age between 18 and 65 years, stable glycemic control(HbA1c ,11% [97 mmol/mol]), and willingness to discon-tinue neuropathic pain medications before MRI scan.Exclusion criteria were clinical evidence of disease in theCNS (e.g., cerebrovascular disease), nondiabetic neuropa-thies, history of alcohol consumption of .20 units/week(1 unit is equivalent to 1 glass of wine or 1 measure ofspirits), diabetic neuropathies other than DSP (e.g., mono-neuropathies, proximal motor neuropathies), epilepsy, re-current severe hypoglycemia, hypoglycemic unawareness,psychiatric conditions, or claustrophobia or other factorsthat preclude MRI. We also recruited age- and sex-matched, right-handed, HVs without diabetes. All HVswere free of chronic pain conditions and were not usinganalgesic medications or alternative therapies for pain.Written informed consent was obtained before participa-tion in the study, which had prior approval by the Shef-field local research ethics committee (reference number08/H1308/276).

Sensory PhenotypingAll subjects underwent detailed clinical and neurophys-iological assessments. The outcome of a detailed upper-and lower-limb neurological examination was gradedusing the Neuropathy Impairment Score questionnaire(20). The following neurophysiological assessments werethen performed: 1) vibration and warmth thermal paindetection thresholds acquired from the dorsal aspectof the right foot using the CASE IV system (WR Med-ical Electronics, Maplewood, MN) using standard tech-niques (21,22), 2) cardiac autonomic function testsperformed with a computer-assisted technique accordingto O’Brien’s protocol (23), and 3) nerve conduction studiesperformed at a stable skin temperature of 31°C anda room temperature of 24°C using a Medelec Synergyelectrophysiological system (Oxford Instruments, Ox-ford, U.K.). The following nerve attributes were mea-sured: 1) sural sensory nerve action potentials andconduction velocities and 2) common peroneal and tibialmotor nerve distal latency, compound muscle actionpotential, and conduction velocity. An overall neuropathycomposite score (NCS) derived from transformed percen-tile points of abnormalities in nerve conduction studies,vibration detection thresholds, and heart rate variabilitywith deep breathing was calculated in accordance with

criteria proposed by Dyck et al. (24). On the basis of theToronto consensus statement (23), all subjects with DSPhad confirmed neuropathy with at least two abnormal-ities on neurophysiological assessments.

Painful DSP Sensate Versus InsensatePainful DSP was defined according to International Associ-ation for the Study of Pain definition of neuropathic pain(25) and scored using the Neuropathy Total Symptom Score-6 (NTSS-6) questionnaire (26). NTSS-6 evaluates the fre-quency and intensity of neuropathic sensory symptomsfrequently reported in DSP (i.e., numbness and/or insensi-tivity, prickling and/or tingling sensation, burning sensation,aching pain and/or tightness; sharp, shooting, lancinatingpain; allodynia and/or hyperalgesia). Only subjects withpainful DSP with distal symmetrical painful neuropathicsymptoms involving the feet and legs present for at least6 months were included.

On the basis of clinical and neurophysiological assess-ments subjects with painful DSP were divided into twogroups: painful DSP insensate (n = 8), wherein subjectsare characterized by lower-limb sensory loss dominatedby small- and large-fiber clinical and neurophysiologicalabnormalities, likely representing the deafferentation orpainful hypoesthesia phenotype (27), and painful DSPsensate (n = 9), wherein subjects have with relativelypreserved small- and/or large-fiber function in combina-tion with thermal (heat) hyperalgesia and low-intensitydynamic mechanical allodynia, likely representing thesensory gain phenotype or irritable nociceptor sub-group described by others (24). The remaining subjectswith diabetes were divided into two control groups (28):no DSP (n = 9), wherein asymptomatic subjects havenormal clinical and neurophysiological assessments,and painless DSP (n = 9), wherein subjects have pain-less neuropathy with abnormal clinical assessmentand at least two abnormalities on neurophysiologicalassessment.

PsychophysicsBefore MRI, we performed a pain calibration procedureusing methods described in previous work (29) (Supple-mentary Data). For each subject, we determined the tem-perature required to obtain a pain score of at least 7 usinga numeric rating scale (NRS) (0 = no pain, 10 = worse painimaginable). A maximum temperature of 47.9°C was usedto prevent scalding.

MRI Acquisition and fMRI Task DesignWe used a 3.0 T scanner (Achieva 3T; Philips MedicalSystems, Holland, the Netherlands) to scan all subjects.Anatomical data were acquired using a T1-weighted mag-netization prepared rapid acquisition gradient echo se-quence with the following parameters: repetition time7.2 ms, echo time 3.2 ms, flip angle 8°, and voxel size0.9 mm3, yielding isotropic spatial resolution. Functionaldata on the basis of the blood oxygenation level–dependent

diabetes.diabetesjournals.org Selvarajah and Associates 797

(BOLD) signal were acquired using a single-shot gradi-ent echo-planar T2*-weighted pulse sequence, with thefollowing parameters: repetition time 3,000 ms, echo time35 ms, and in-plane pixel dimensions 1.8 3 1.8 mm. Con-tiguous transaxial slices with slice thickness of 4 mm wereoriented in the oblique axial plane parallel to the anterior-posterior commissure bisection, covering the whole cortex,with partial coverage of the cerebellum. Two hundred threetemporal dynamics were acquired per boxcar functionrun. Throughout scanning, we also continuously mon-itored heart rate (electrocardiogram and peripheralpulse) and respiration through the built-in scannermonitoring/triggering device (Philips Medical Systems).

After calibration and training, subjects were positionedin the scanner, and an MRI-compatible heat probe wasplace on the dorsum of the right foot or anterior aspect ofthe thigh. Subjects were randomized to begin stimulationof either the foot or the thigh. During each period, fMRIimages were acquired over seven functional runs. A classicblock design was used consisting of baseline and stimulus,alternating seven times. Each functional run comprised analternating 30-s period of tonic heat stimulation followedby rest. The intensity of painful heat stimulation appliedto the foot and thigh was calibrated for each subject. Theslope of thermal stimulus applied was 5°C/s. Thermal stim-ulus was delivered using a Peltier-driven thermotest devicewith an fMRI-compatible thermode (probe size 3 3 3 cm,PATHWAY Model CHEPS neurosensory analyzer; Medoc,Ramat Yishai, Israel). The duration of rest periods waspseudorandomized to 50, 55, and 60 s to minimize theanticipation of the next stimulation period (Fig. 1 andSupplementary Data). Subjects were instructed to keep stillduring the scanning, which was verified visually. At the endof the scan, subjects once again were asked to rate the level

of heat pain in the foot and thigh experienced during thefMRI. We also asked subjects to rate the unpleasantness ofthe experience when heat pain was applied to the foot andthigh (separately) using an NRS (0 = not unpleasant/tolerable,10 = extremely unpleasant/intolerable).

MRI Analyses

Cortical ThicknessCortical reconstruction and volumetric segmentation wereperformed with FreeSurfer software (http://surfer.nmr.mgh.harvard.edu). This processing includes motion correc-tion and averaging (30) of volumetric T1-weighted images,removal of nonbrain tissue using a hybrid watershed/surface deformation procedure (31), affine registration to theTalairach atlas (32,33), intensity normalization, tessellationof the gray matter-white matter boundary, automatedtopology correction (34,35), and surface deformation fol-lowing intensity gradients to optimally place the gray/whiteand gray/cerebrospinal fluid borders at the location wherethe greatest shift in intensity defines the transition to theother tissue class (36,37). Surface-based maps are createdfrom the intensity and continuity information from theentire three-dimensional MR volume in segmentation anddeformation procedures to produce representations of cor-tical thickness calculated as the closest distance from thegray/white boundary to the gray/cerebrospinal fluid bound-ary at each vertex on the tessellated surface (34). Corticalthickness (in mm, sensorimotor regions S1 and precentralcortex) and deep brain nuclei volumes (in mm3, thalamus,caudate, and insula) were measured. Results of corticalthickness analyses are presented in Table 3 and Fig. 2.Significant differences in regional cortical thickness anddeep brain nuclei volumes among study cohorts were ex-amined using ANOVA. In addition, S1 cortical thickness

Figure 1—Sensory descriptors. Means and 95%CIs of the six pain descriptors and total score from the NTSS-6 by subjects with painful DSPsensate and insensate. *P , 0.05.

798 Somatosensory Cortex Abnormalities in DSP Diabetes Volume 68, April 2019

that significantly correlated to the degree of neuropathyseverity (NCS), and pain scores (NTSS-6 and evoked pain)were determined in subjects with painful DSP (sensate andinsensate) using Pearson correlation for normally distributeddata and Spearman rank correlation for nonnormally dis-tributed data. This analysis was used to examine the asso-ciation between volume changes in the S1 and markers ofneuropathy severity.

fMRI AnalysisMRI data acquisition, preprocessing (FNIRT), and analysesfollowed standard procedures (Supplementary Data). Func-tional imaging data were processed using the Oxford Centrefor Functional MRI of the Brain (FMRIB) expert analysistool FEAT version 5.98 (www.fmrib.ox.ac.uk/fsl) (38). Toallow us to focus specifically on topographical shifts in bodypart representation, we conducted two different analyses.

Cortical Distance Analysis of Peak Activation. First, weexamined differences in the Euclidean distances (EDs) ofpeak activation within the contralateral S1. A region ofinterest (ROI) comprising the contralateral S1 was used torestrict the fMRI analysis to the gray matter of the S1 foreach subject. The maximally activated voxel within thisROI during each tonic heat stimulation paradigm wasdetermined. The spatial position of this voxel was locatedfor each subject by measuring the ED between it anda standard anatomical point (the point at which the centralsulcus meets the longitudinal fissure at the dorsal aspect ofthe brain). This procedure has been described in detailpreviously (13). The ED between the anatomical markerand the maximally activated voxel in the S1 ROI wascomputed for the anterior-posterior, medial-lateral, andsuperior-inferior coordinates. The ED between two pointswas calculated using the Pythagorean theorem and pro-vided an absolute value independently of direction (12). AnANOVA was used to determine the significant differencesin the ED of peak S1 activation between study cohorts.However, the ED method only examines the differences inthe location of peak activation during thermal nociceptivestimulation between cohorts; it does not examine the fullextent of neuronal activation.

Analysis of the Extent of S1 Receptive Field Activa-tion. Next, to examine the group differences in the extentof S1 activation, we estimated activity maps for eachcondition (foot and thigh stimulation) within each sub-ject by applying a voxel-based general linear model, asimplemented in FEAT. For each subject, a set of regres-sors was constructed for conditions of interest (heatstimulation of the foot and thigh) using the block designparadigm convolved with a g-function, and its temporalderivative was used to model the activation time course.The parameter estimates (regression slopes) for eachcondition thus provided an estimate at each voxel ofthe activation intensity for that condition. The second-level analysis (i.e., group analysis) was performed toidentify condition-specific patterns of activation foreach group (i.e., HV, no DSP, painless DSP, painful

DSP sensate, painful DSP insensate). This whole-braingroup-level analysis was carried out using FMRIB’s localanalysis of mixed effects (39). The main condition con-trasts were defined using painful DSP insensate as themain comparator group versus 1) HV, 2) no DSP, 3) pain-less DSP, and 4) painful DSP sensate. Z (GaussianizedT/F) statistic images were thresholded using clustersdetermined by Z .2, and a family-wise error–correctedcluster significance threshold of P , 0.05 was applied tothe suprathreshold clusters to correct for multiple compar-isons. For presentation purposes, the activation maps weresmoothed using a full-width at half maximum of 4 mmand thresholded at Z .1.5. A group-level analysis was per-formed using FMRIB’s local analysis of mixed effects, withpainful DSP insensate as the main comparator group. Usingthis analysis, we compared the pattern of neuronal activa-tion in subjects with painful DSP insensate with the otherstudy cohorts.We asked which brain regions aremore activeduring nociceptive warm thermal stimulation in subjectswith painful DSP insensate compared with the othercohorts and vice versa. The resulting maps were visuallyinspected to verify that the clusters of activation werelocated along the S1 strip. For each comparison, we iden-tified voxels within the S1 that displayed significantincreases in signal intensity. In each subject, we determinedthe percent change during nociceptive warm thermal stim-ulation (relative to baseline rest periods) in signal intensityover time centered around the maximally activated voxeland the significant differences between groups assessed(Mann-Whitney U test P , 0.05). In addition, Z scoresof S1 signal intensity that were significantly correlated toseverity of neuropathy (NCS), pain scores (NTSS-6 andevoked pain scores), and S1 cortical thickness were de-termined using Pearson correlation for normally distributeddata and Spearman rank correlation for nonnormally dis-tributed data. This analysis was performed to examine theassociation between the functional reorganization of the S1and S1 cortical volume changes and clinical markers ofneuropathy severity. The Z score was chosen because it isassumed to be more appropriate than the magnitude ofdifference in that it also considers the variance in the signal.

Statistical AnalysesStatistical analysis was carried out with SPSS version 21software (IBM Corporation). Group differences on demo-graphic characteristics and psychophysical experimentswere compared using ANOVA. Subsequent subgroup com-parisons were performed using a post hoc analysis.

RESULTS

Overall, subjects with DSP were older (F[4,39] = 0.36,ANOVA P = 0.83) than subjects with no DSP. To minimizethe potential effect of age on study outcome measures, werecruited an older cohort of HVs. No significant differencewas found in duration of diabetes between the diabetesgroups (F[3,31] = 0.53, ANOVA P = 0.98) (Table 1). Therewere no significant differences in post hoc analyses of

diabetes.diabetesjournals.org Selvarajah and Associates 799

group mean for age or duration of diabetes, and there wasno significant difference in mean HbA1c (F[3,31] = 1.65,ANOVA P = 0.20). As expected, a x2 test for independenceindicated a significant association between neuropathyand retinopathy status (x2[6,35] = 13.8, P = 0.03).

Sensory ProfilingAll subjects with painful DSP had a distal symmetricalperipheral neuropathy that affected the feet and legs and wassparing of the thigh. Pain duration (t[15] = 0.32, P = 0.76) andintensity (t[15] = 0.14, P = 0.89) was comparable between thetwo painful DSP groups (Table 1). Using the NTSS-6 to assesspain descriptors, it was found that subjects with painful DSPreported typical neuropathic symptoms (Fig. 1). Those withpainful DSP sensate, however, reported significantly highermean scores for allodynia than those with painful DSP in-sensate (2.48 [1.4] vs. 0.83 [1.5], t[15] = 2.27, P = 0.04).

PsychophysicsNo significant difference in temperature was applied to thefoot (F[4,39] = 1.3, P = 0.29) (Table 2 includes post hocanalyses). Unsurprisingly, subjects with painless DSP andpainful DSP insensate had the lowest NRS pain scores(F[4,39] = 12.4, ANOVA P , 0.001) and unpleasantnessscores (F[4,39] = 2.92, ANOVA P = 0.03). Upon thighstimulation, no significant difference was found in tem-perature (F[4,39] = 1.03, ANOVA P = 0.40), NRS pain(F[4,39] = 0.59, ANOVA P = 0.67), or unpleasantness(F[4,39] = 0.24, ANOVA P = 0.92) scores.

Cortical ThicknessSubjects with painful DSP insensate had significantly lowermean bilateral S1 (3.60 [0.1] mm, F[4,39] = 5.78, ANOVAP = 0.001) cortical thickness compared with other studycohorts (HV 4.04 [0.1] mm, no DSP 3.85 [0.2] mm,

Table 1—Demographic characteristics of each study cohort

HV No DSP Painless DSPPainful DSPsensate Painful DSP insensate P value

n 9 9 9 9 8

Age (years) 51.5 (7.9) 45.9 (10.1) 46.3 (12.1) 48.4 (12.0) 44.5 (12.1) 0.83

Sex, male (n) 1 6 7 5 6 0.03

Diabetes duration (years) NA 22.8 (14.9) 21.4 (8.1) 23.4 (10.2) 23.2 (12.6) 0.98

Pain duration (years) NA NA NA 6.8 (6.3) 7.8 (7.3) 0.76

HbA1c (mmol/mol) NA 72.8 (17.4) 68.0 (7.7) 72.8 (13.9) 87.7 (28.1) 0.30

HbA1c (%) NA 8.8 8.4 8.8 10.2

NTSS-6 score NA NA NA 11.3 (5.1) 11.2 (4.4) 0.96

Vibration JND NA 14.7 (2.2) 19.3 (4.6) 21.4 (3.9) 24.3 (1.7) 0.002

NCS NA 1.4 (1.8) 10.6 (7.1) 9.0 (5.3) 19.1 (0.83) <0.001

Retinopathy status (n) NA 0.03No DR 2 3 2 2Mild nonproliferative DR 7 3 4 0Moderate/severe nonproliferative DR 0 3 3 6

Data are mean (SD) unless otherwise indicated. ANOVA was used to determine the P value. Boldface type indicates significance at P ,0.05. DR, diabetic retinopathy; JND, just noticeable difference; NA, not applicable.

Table 2—Results of psychophysical experiments conducted before MRI

HV No DSP Painless DSP Painful DSP sensate Painful DSP insensate P value

FootTemperature 45.7 (1.1) 46.2 (1.3) 47.1 (1.0) 46.5 (1.1) 46.1 (1.7) 0.29Pain score 8.1 (0.8) 8.1 (0.4) 5.4 (3.5) 7.4 (1.6) 1.6 (3.1) <0.001Unpleasantness score 7.2 (1.6) 7.1 (1.3) 5.6 (3.4) 5.5 (2.1) 3.4 (3.3) 0.03

ThighTemperature 45.7 (1.4) 45.5 (1.2) 45.3 (1.6) 46.5 (0.7) 45.8 (1.4) 0.40Pain score 7.5 (0.8) 7.7 (0.8) 7.9 (0.9) 7.1 (1.4) 7.5 (1.9) 0.59Unpleasantness score 6.2 (2.3) 6.6 (1.4) 6.0 (2.8) 6.1 (1.8) 6.2 (2.3) 0.92

Data are mean (SD). Boldface type indicates significance at P , 0.05. Noxious thermal heat stimulation was applied to the right foot(neuropathic site) and thigh (nonneuropathic control site). Mean temperature required to achieve a pain score (0 = no pain, 10 =worst pain)of at least 7 was recorded. Subjects also were asked to rate unpleasantness (0 = not unpleasant, 10 = most unpleasant). Sequence ofthermal stimulus application to foot and thigh was randomized across subjects. Post hoc analysis for foot pain score: painful DSPinsensate vs. HV (P, 0.001), no DSP (P, 0.001), and painful DSP sensate (P, 0.001); painless DSP vs. HV (P = 0.02), no DSP (P = 0.02),and painful DSP sensate (P = 0.06); and painful DSP insensate vs. painless DSP (P = 0.002). Post hoc analysis for foot unpleasantnessscore: painful DSP insensate vs. HV (P = 0.005) and no DSP (P = 0.006).

800 Somatosensory Cortex Abnormalities in DSP Diabetes Volume 68, April 2019

painless DSP 3.82 [0.2] mm, painful DSP sensate 3.88 [0.2]mm) (Table 3 [includes post hoc analyses] and Fig. 2).Unilateral evaluation showed significant differences inboth left- and right-side S1 cortical thickness (F[4,39] =5.30 [ANOVA P = 0.002] and 6.46 [ANOVA P , 0.001],respectively). On average, somatosensory cortical thick-ness was 12.2% lower in subjects with painful DSPinsensate than in HVs. We also examined other pain,sensory, and motor processing areas, and the only othersignificant difference was in the precentral gyrus thickness(painful DSP insensate 4.43 [0.2] mm vs. HV 4.73 [0.2]mm, no DSP 4.74 [0.3] mm, painless DSP 4.63 [0.2] mm,painful DSP sensate 4.77 [0.2] mm; F[4,39] = 4.98; ANOVAP = 0.009). No significant difference emerged betweenpainful DSP and other study cohorts in bilateral measuresfor caudate volume (F[4,39] = 0.92, ANOVA P . 0.4), anddifferences fell short of significance for thalamic volume(F[4,39] = 1.83, ANOVA P = 0.14) and insula corticalthickness (F[4,39] = 1.90, ANOVA P = 0.13). Unilateralevaluation showed no significant differences in either left-or right-sided caudate and a trend-level difference in leftthalamus (F[4,39] = 2.17, ANOVA P = 0.09). For subjectswith painful DSP (sensate and insensate), a moderate andsignificant correlation was found between foot-evoked painscores and S1 cortical thickness but not for the precentralgyrus thickness (Fig. 2). Evoked foot pain scores, a measureof small c-fiber function and foot sensitivity, correlated withleft-side (r = 0.44, P = 0.004) and right-side (r = 0.47, P =0.002) S1 cortical thickness (Fig. 2C). Furthermore, therewas a significant negative correlation between NCS and S1cortical thickness (r = 20.60, P = 0.01) (Fig. 2D). Nosignificant correlations were found between thigh-evokedpain scores and measures of cortical thickness. For subjectswith painful DSP (sensate and insensate), there was nosignificant correlation between reported pain scores (NTSS-6) and measures of cortical thickness or volumes of deep

gray matter nuclei. There was no significant difference inperipheral gray matter (F[4,39] = 1.2, ANOVA P = 0.32),total gray matter (F[4,39] = 0.85, ANOVA P = 0.50), andwhole-brain (F[4,39] = 2.1, ANOVA P = 0.11) volumesamong study groups.

fMRIBecause subjects with painful DSP insensate had the great-est reduction in S1 cortical thickness, we reasoned thatthere will be accompanying functional changes in thetopographical organization of the S1. Statistical paramet-ric maps showing brain areas significantly activated inresponse to stimulation are shown in SupplementaryFig. 3. In all groups, tonic heat stimulation of the right foot(Supplementary Fig. 3A) resulted in increased BOLD re-sponse in the left-side S1, thalamus, insula, and anteriorcingulate gyrus consistent with the well-described repre-sentation of pain. Similar areas of increased BOLD re-sponse were seen with thigh stimulation (SupplementaryFig. 3B). In all subjects, tonic heat stimulation of the footand thigh resulted in a lateral-to-medial pattern of acti-vation within the contralateral S1, consistent with thesensory homunculus. The precise locations of S1 acti-vations were similar in all four groups. The ED of peakactivation within the left-side S1 was not significantlydifferent across study groups for both foot (F[4,39] =0.64, ANOVA P = 0.63) (Supplementary Fig. 2) and thigh(F[4,39] = 0.39, ANOVA P = 0.85) stimulation.

There were, however, significant differences in BOLDresponse in subjects with painful DSP insensate comparedwith the other study cohorts during both foot (Fig. 3) andthigh (Fig. 4) stimulation. That is, subjects with painfulDSP insensate displayed significant S1 functional reorga-nization in which there was an expansion of the arearepresenting pain in the lower limb to include regionsrepresenting the face and lips. Finally, there was

Table 3—Brain volumes, bilateral regional cortical thicknesses, and deep brain nuclei volumes

HV No DSP Painless DSPPainful DSPsensate

Painful DSPinsensate P value

Brain volume, mLPeripheral gray 626.7 (27.2) 622.6 (24.6) 604.9 (23.8) 606.8 (52.4) 591.6 (25.0) 0.32Gray matter 792.9 (30.9) 788.7 (31.8) 767.9 (29.0) 763.6 (62.5) 777.5 (43.5) 0.50White matter 729.8 (41.4) 721.0 (43.9) 709.8 (27.6) 694.5 (26.7) 751.3 (12.9) 0.04Whole brain 1,522.7 (59.1) 1,509.6 (64.7) 1,477.7 (27.1) 1,458.0 (83.4) 1,528.9 (46.0) 0.11

Cortical thickness, mmPostcentral 4.04 (0.1) 3.85 (0.2) 3.82 (0.2) 3.88 (0.2) 3.60 (0.1) 0.001Precentral 4.73 (0.2) 4.74 (0.3) 4.63 (0.2) 4.77 (0.2) 4.43 (0.2) 0.009

Deep brain nuclei volume, cm3

Thalamus 6.55 (0.5) 7.39 (0.8) 6.91 (0.4) 6.66 (0.9) 6.71 (0.8) 0.14Caudate 3.40 (0.5) 3.64 (0.5) 3.67 (0.4) 3.41 (0.5) 3.68 (0.3) 0.46Insula 5.81 (0.3) 5.89 (0.2) 5.90 (0.3) 5.71 (0.4) 5.56 (0.2) 0.13

Data are mean (SD). Boldface type indicates significance at P, 0.05. Post hoc analysis: white matter volume: painful DSP insensate vs.painless DSP (P = 0.03), painful DSP insensate vs. painful DSP sensate (P = 0.004), and HV vs. painful DSP sensate (P = 0.04); postcentralgyrus thickness: painful DSP insensate vs. HV (P , 0.001), no DSP (P = 0.008), painless DSP (P = 0.02), and painful DSP sensate (P =0.003); precentral gyrus thickness: painful DSP insensate vs. HV (P = 0.004), no DSP (P = 0.003), painless DSP (P = 0.05), and painful DSPsensate (P = 0.002).

diabetes.diabetesjournals.org Selvarajah and Associates 801

a significant relationship between the activation intensity inthe face/lip region and severity of neuropathy (NCS r = 0.74,P, 0.001), neuropathic pain score (NTSS-6 r =20.45, P =0.04), evoked foot pain score (r = 20.46, P = 0.03), and S1cortical thickness (r = 20.57, P = 0.01) (SupplementaryFig. 4).

DISCUSSION

Using a multimodal neuroimaging approach, these datashow a reduction in S1 cortical thickness and a remappingof S1 sensory processing in patients with painful DSPinsensate and that the magnitude of cortical response inthe face/lip S1 region to tonic heat stimulation is nega-tively correlated with S1 cortical thickness. Furthermore,the extent to which S1 structure and function is altered isrelated to the severity of neuropathy and the magnitude ofself-reported and tonic heat-evoked pain intensity ratings.Thus, only in the painful DSP insensate group is thealtered structural and functional organization of S1linked to both behavioral and brain responses to heathyperalgesia. This study is the first in our knowledge todemonstrate a relationship between patient sensory phe-notypes and brain structural and functional changes inDSP.

Although the painful/painless diabetic foot has longbeen recognized, there has been a lack of a pathophysi-ological explanation for such patients who have severeneuropathic pain in the absence of sensation in their feet(40). Advances have been made in the sensory profiling ofpatients with diabetes (Pain in Neuropathy Study [PiNS])(4) as well as early indications that an individual’s painphenotype may predict response to treatment (41). How-ever, a pathophysiological explanation at both the periph-eral and the central level for these phenotypes is limited.The current study has tried to explore in carefully pheno-typed patients whether a correlation exists between CNSstructural/functional changes and various sensory pheno-types. Patients with painful DSP insensate had the greatestseverity of neuropathy (NCS) and the greatest reduction inS1 cortical thickness. Hence, the most likely explanationfor the reduction in S1 cortical thickness is deafferentationor a dying-back axonopathy that principally affects thesensory neurons in DSP. This explanation is in keepingwith our previous findings (5,15,16,42,43), but we alsodemonstrate a widening of S1 functional representationof both the foot and the thigh in patients with painfulDSP insensate. It is plausible that the reduction incortical thickness has led to recruitment of viable/functioning

Figure 2—A and B: Box-and-whisker plots (median and quartiles) of postcentral gyrus and precentral gyrus cortical thickness (in mm) foreach study cohort. C and D: Spearman rank correlation between S1 cortical thickness (in mm) and foot pain score and NCS S1 corticalthickness for subjects with painful (P) DSP sensate and P-DSP insensate.

802 Somatosensory Cortex Abnormalities in DSP Diabetes Volume 68, April 2019

Figure 3—Whole-brain voxel-wise group analysis of tonic heat stimulation of the foot for subjects with painful DSP insensate compared withHVs (A andB) and subjects with no DSP (C andD), painless DSP (E and F ), and painful DSP sensate (G andH). Themap shows activation thatexceeds a threshold of Z.1.5 and a family-wise error–corrected cluster significance threshold of P, 0.05 (for display only). A, C, E, and G:The blue circle demonstrates the region of greatest activation within the S1 in HVs and subjects with no DSP, painless DSP, and painful DSPsensate compared with subjects with painful DSP insensate. B, D, F, and H: The blue circle demonstrates the region of greatest activationwithin the S1 in subjects with painful DSP insensate compared with the other study cohorts. Box plots display median, interquartile range,and minimum and maximum values for percent change in signal intensity (SI%) from the region of greatest activation highlighted by the bluecircle. Subjects with painful DSP insensate showed displacement of the lateral border of the foot into the face/lip area. Group comparisons ofparameter estimates were performed using Mann-Whitney U test, and P , 0.05 (two-tailed) indicated significance.

diabetes.diabetesjournals.org Selvarajah and Associates 803

Figure 4—Within-group analysis of tonic heat stimulation of the thigh for subjects with painful DSP insensate compared with HVs (A and B)and subjects with no DSP (C and D), painless DSP (E and F), and painful DSP sensate (G andH). A,C, E, andG: The blue circle demonstratesthe region of greatest activation within the S1 in HVs and subjects with no DSP, painless DSP, and painful DSP sensate compared withsubjects with painful DSP insensate. B, D, F, and H: The blue circle demonstrates the region of greatest activation within the S1 in subjectswith painful insensate compared with the other study cohorts. Box plots display median, interquartile range, and minimum and maximumvalues for percent change in signal intensity (SI%) from the region of greatest activation highlighted by the blue circle. Subjects with painfulDSP insensate showed displacement of the lateral border of the thigh into the face/lip area. Group comparisons of parameter estimates wereperformed using Mann-Whitney U test, and P , 0.05 (two-tailed) indicated significance.

804 Somatosensory Cortex Abnormalities in DSP Diabetes Volume 68, April 2019

neurons in adjacent areas, which is similar to findingsreported in previous studies of amputees with phantomlimb pain (44). Of note, there was a negative relationshipbetween the magnitude of cortical response in the face/lipregion to tonic heat stimulation with S1 cortical thicknessand both evoked pain and chronic neuropathic painseverity, suggesting that reduction in cortical thicknessresults in fewer viable neurons and less overall painperception. In addition, further studies should examinethe role of pain as a cause and/or consequence of the CNSchanges described. Another possible explanation thatneeds future exploration is the impact of diabetes statuson these findings. S1 cortical thickness was lower in thediabetes groups compared with HVs, which would suggestthat diabetes also affects the brain structural changesdescribed. Finally, although our findings suggest thatthe changes in S1 structural and functional organization isdriven by the neuropathic process, other pain-relatedbehavioral changes, such as physical functioning, depriva-tion of social contacts, and mood, also may contributeto brain neuroplasticity. Our study was not designed toexplore the interaction of these various factors on brainchanges in chronic pain. Furthermore, prospective studiesin well-characterized patients will be required to examinethis.

This study is limited by its relatively small sample size;however, the subjects were well characterized, and webelieve that the study provides a novel insight into theCNS neural correlates of various clinical pain phenotypesin DSP. The results obtained are robust (P , 0.05), evenafter correction for multiple comparisons using the family-wise error method. In addition, we examined only subjectswith type 1 diabetes. Although the pathophysiology ofnerve injury in type 2 diabetes may be different, wepostulate similar alterations within the CNS. Nevertheless,further studies in subjects with type 2 diabetes are war-ranted. There was also a significant difference in sexdistribution across study cohorts, but this is unlikely tohave a significant impact on the assessments because MRIimages were registered to a standard space that removesglobal differences in cortical morphometry that may berelated to sex, hemispheric asymmetries, or age. Further-more, no significant differences were found between malesand females in the psychophysical responses to nociceptivestimulation to the foot and thigh (temperature, pain, andunpleasantness scores). Hence, subgroup differences in sexdistribution is unlikely to affect the functional activationof the somatosensory cortex. Finally, apart from the HVcohort, there was a male preponderance in all the other(diabetes) subgroups, including the disease control groupcomprising subjects with diabetes and no DSP. No statis-tically significant differences were present between controlgroups (HV vs. no DSP). Nevertheless, important sexdifferences exist in the prevalence of painful diabeticneuropathy and other chronic pain conditions. A largerstudy with a different design is required to truly examinesex differences in functional organization of the S1.

Several critical lines of future work have emerged fromthis study, including longitudinal prospective studies todetermine the natural history of brain structural andfunctional changes in DSP and to truly dissect whetherand how these are causally related.

Acknowledgments. The authors acknowledge the hard work, skills, andcontributions of the radiographers at the Academic Unit of Radiology. The authorsalso greatly appreciate the study volunteers who spent considerable timeparticipating in this study.Funding. This study was funded by a grant from JDRF International (1-2008-69).Duality of Interest. No potential conflicts of interest relevant to this articlewere reported.Author Contributions. D.S. recruited participants, undertook clinical andneurophysiological assessments, researched and analyzed clinical and MR data,and wrote the manuscript. I.D.W. undertook planning of MR experiments andreviewed and revised the manuscript. F.F. analyzed MR cortical thickness data.A.S., J.D., and E.B. undertook clinical and neurophysiological assessments andassisted with participant recruitment. J.H. undertook clinical and neurophysio-logical assessments, assisted with participant recruitment, and assisted withanalyzing MR data. G.R. supervised neurophysiological assessments and reviewedand revised the manuscript. R.G. reviewed and revised the manuscript. I.T.supervised MRI experiments, assisted with data analysis and interpretation, andreviewed and revised the manuscript. S.T. supervised the project and reviewedand revised the manuscript. S.T. is the guarantor of this work and, as such, had fullaccess to all the data in the study and takes responsibility for the integrity of thedata and the accuracy of the data analysis.Data Availability. The data sets generated and/or analyzed during the studyare available from the corresponding author upon reasonable request.

References1. Dyck PJ, Kratz KM, Karnes JL, et al. The prevalence by staged severityof various types of diabetic neuropathy, retinopathy, and nephropathy ina population-based cohort: the Rochester diabetic neuropathy study. Neurology1993;43:817–8242. Tesfaye S, Boulton AJ, Dickenson AH. Mechanisms and management of diabeticpainful distal symmetrical polyneuropathy. Diabetes Care 2013;36:2456–24653. Feldman EL, Nave KA, Jensen TS, Bennett DLH. New horizons in diabeticneuropathy: mechanisms, bioenergetics, and pain. Neuron 2017;93:1296–13134. Themistocleous AC, Ramirez JD, Shillo PR, et al. The Pain in NeuropathyStudy (PiNS): a cross-sectional observational study determining the somato-sensory phenotype of painful and painless diabetic neuropathy. Pain 2016;157:1132–11455. Selvarajah D, Wilkinson ID, Davies J, Gandhi R, Tesfaye S. Central nervoussystem involvement in diabetic neuropathy. Curr Diab Rep 2011;11:310–3226. Marshall AG, Lee-Kubli C, Azmi S, et al. Spinal disinhibition in experimentaland clinical painful diabetic neuropathy. Diabetes 2017;66:1380–13907. Segerdahl AR, Themistocleous AC, Fido D, Bennett DL, Tracey I. A brain-based pain facilitation mechanism contributes to painful diabetic polyneuropathy.Brain 2018;141:357–3648. Lee MC, Tracey I. Imaging pain: a potent means for investigating painmechanisms in patients. Br J Anaesth 2013;111:64–729. Merzenich MM, Nelson RJ, Stryker MP, Cynader MS, Schoppmann A, ZookJM. Somatosensory cortical map changes following digit amputation in adultmonkeys. J Comp Neurol 1984;224:591–60510. Birbaumer N, Lutzenberger W, Montoya P, et al. Effects of regional anes-thesia on phantom limb pain are mirrored in changes in cortical reorganization. JNeurosci 1997;17:5503–550811. Flor H, Elbert T, Knecht S, et al. Phantom-limb pain as a perceptual correlateof cortical reorganization following arm amputation. Nature 1995;375:482–484

diabetes.diabetesjournals.org Selvarajah and Associates 805

12. Wrigley PJ, Press SR, Gustin SM, et al. Neuropathic pain and primary so-matosensory cortex reorganization following spinal cord injury. Pain 2009;141:52–5913. Lotze M, Flor H, Grodd W, Larbig W, Birbaumer N. Phantom movements andpain. An fMRI study in upper limb amputees. Brain 2001;124:2268–227714. Eaton SE, Harris ND, Rajbhandari SM, et al. Spinal-cord involvement indiabetic peripheral neuropathy. Lancet 2001;358:35–3615. Selvarajah D, Wilkinson ID, Emery CJ, et al. Early involvement of the spinalcord in diabetic peripheral neuropathy. Diabetes Care 2006;29:2664–266916. Selvarajah D, Wilkinson ID, Maxwell M, et al. Magnetic resonance neuro-imaging study of brain structural differences in diabetic peripheral neuropathy.Diabetes Care 2014;37:1681–168817. Apkarian AV, Sosa Y, Sonty S, et al. Chronic back pain is associated withdecreased prefrontal and thalamic gray matter density. J Neurosci 2004;24:10410–1041518. Schaechter JD, Moore CI, Connell BD, Rosen BR, Dijkhuizen RM. Structuraland functional plasticity in the somatosensory cortex of chronic stroke patients.Brain 2006;129:2722–273319. DaSilva AF, Becerra L, Pendse G, Chizh B, Tully S, Borsook D. Colocalizedstructural and functional changes in the cortex of patients with trigeminal neu-ropathic pain. PLoS One 2008;3:e3396.20. Bril V. NIS-LL: the primary measurement scale for clinical trial endpoints indiabetic peripheral neuropathy. Eur Neurol 1999;41(Suppl. 1):8–1321. Dyck PJ, O’Brien PC, Kosanke JL, Gillen DA, Karnes JLA. A 4, 2, and1 stepping algorithm for quick and accurate estimation of cutaneous sensationthreshold. Neurology 1993;43:1508–151222. Dyck PJ, Zimmerman I, Gillen DA, Johnson D, Karnes JL, O’Brien PC. Cool,warm, and heat-pain detection thresholds: testing methods and inferences aboutanatomic distribution of receptors. Neurology 1993;43:1500–150823. Pfeifer M. Cardiovascular assessment. In Diabetic Neuropathy. 2nd ed. DyckPJ, Thomas PK, Eds. Philadelphia, Saunders, 1999, p. 171–18324. Dyck PJ, Davies JL, Litchy WJ, O’Brien PC. Longitudinal assessment ofdiabetic polyneuropathy using a composite score in the Rochester DiabeticNeuropathy study cohort. Neurology 1997;49:229–23925. Finnerup NB, Haroutounian S, Kamerman P, et al. Neuropathic pain: anupdated grading system for research and clinical practice. Pain 2016;157:1599–160626. Bastyr EJ III, Price KL, Bril V; MBBQ Study Group. Development and validitytesting of the neuropathy total symptom score-6: questionnaire for the study ofsensory symptoms of diabetic peripheral neuropathy. Clin Ther 2005;27:1278–129427. Maier C, Baron R, Tölle TR, et al. Quantitative sensory testing in the GermanResearch Network on Neuropathic Pain (DFNS): somatosensory abnormalities in1236 patients with different neuropathic pain syndromes. Pain 2010;150:439–45028. Tesfaye S, Boulton AJ, Dyck PJ, et al.; Toronto Diabetic Neuropathy ExpertGroup. Diabetic neuropathies: update on definitions, diagnostic criteria, estimation

of severity, and treatments [published correction appears in Diabetes Care 2010;33:2725]. Diabetes Care 2010;33:2285–229329. Atlas LY, Bolger N, Lindquist MA, Wager TD. Brain mediators of predictive cueeffects on perceived pain. J Neurosci 2010;30:12964–1297730. Reuter M, Rosas HD, Fischl B. Highly accurate inverse consistent registration:a robust approach. Neuroimage 2010;53:1181–119631. Ségonne F, Dale AM, Busa E, et al. A hybrid approach to the skull strippingproblem in MRI. Neuroimage 2004;22:1060–107532. Fischl B, Salat DH, Busa E, et al. Whole brain segmentation: automatedlabeling of neuroanatomical structures in the human brain. Neuron 2002;33:341–35533. Fischl B, Salat DH, van der Kouwe AJ, et al. Sequence-independent seg-mentation of magnetic resonance images. Neuroimage 2004;23(Suppl. 1):S69–S8434. Fischl B, Liu A, Dale AM. Automated manifold surgery: constructing geo-metrically accurate and topologically correct models of the human cerebral cortex.IEEE Trans Med Imaging 2001;20:70–8035. Ségonne F, Pacheco J, Fischl B. Geometrically accurate topology-correctionof cortical surfaces using nonseparating loops. IEEE Trans Med Imaging 2007;26:518–52936. Dale AM, Fischl B, Sereno MI. Cortical surface-based analysis. I. Seg-mentation and surface reconstruction. Neuroimage 1999;9:179–19437. Fischl B, Dale AM. Measuring the thickness of the human cerebral cortexfrom magnetic resonance images. Proc Natl Acad Sci U S A 2000;97:11050–1105538. Smith SM, Jenkinson M, Woolrich MW, et al. Advances in functional andstructural MR image analysis and implementation as FSL. Neuroimage 2004;23(Suppl. 1):S208–S21939. Winkler AM, Ridgway GR, Webster MA, Smith SM, Nichols TE. Permutationinference for the general linear model. Neuroimage 2014;92:381–39740. Veves A, Manes C, Murray HJ, Young MJ, Boulton AJ. Painful neuropathy andfoot ulceration in diabetic patients. Diabetes Care 1993;16:1187–118941. Demant DT, Lund K, Vollert J, et al. The effect of oxcarbazepine inperipheral neuropathic pain depends on pain phenotype: a randomised,double-blind, placebo-controlled phenotype-stratified study. Pain 2014;155:2263–227342. Selvarajah D, Wilkinson ID, Emery CJ, et al. Thalamic neuronal dysfunctionand chronic sensorimotor distal symmetrical polyneuropathy in patients with type1 diabetes mellitus. Diabetologia 2008;51:2088–209243. Selvarajah D, Wilkinson ID, Gandhi R, Griffiths PD, Tesfaye S. Microvascularperfusion abnormalities of the thalamus in painful but not painless diabeticpolyneuropathy: a clue to the pathogenesis of pain in type 1 diabetes. DiabetesCare 2011b;34:718–72044. Kew JJ, Ridding MC, Rothwell JC, et al. Reorganization of cortical blood flowand transcranial magnetic stimulation maps in human subjects after upper limbamputation. J Neurophysiol 1994;72:2517–2524

806 Somatosensory Cortex Abnormalities in DSP Diabetes Volume 68, April 2019